CVE-2022-34670

What it actually does

The flaw resides in the kernel mode layer handler of the NVIDIA GPU Display Driver for Linux; an unprivileged user can supply input that triggers truncation when casting a primitive to a smaller primitive type, discarding data and enabling denial of service or information disclosure.

Why it matters to the business

Local users could crash GPU-dependent workloads or leak kernel-adjacent data on Linux systems, affecting stability of graphics, compute, and AI jobs; the issue has remained unexploited in the wild for nearly four years with negligible EPSS and no ransomware or KEV association.

How an attack unfolds

An attacker with a local account on an affected Linux system interacts with the NVIDIA kernel driver, submits crafted values that force an unsafe primitive cast, induces truncation, and leverages the resulting memory or state corruption to achieve denial of service or read sensitive information.

How to fix it

Upgrade openfga/openfga (GIT) to commit d33a87af80688acdad1225b23e7be6655b323d35; organizations should also apply the corresponding NVIDIA Linux driver update that resolves the truncation logic in the kernel handler.

How to confirm the fix worked

Confirm the installed driver or package reports the fixed commit hash d33a87af80688acdad1225b23e7be6655b323d35; recompile or reinstall the module and verify that the kernel handler no longer permits the previously successful truncation test cases.

Typical attack path

1. Obtain a local unprivileged shell on a Linux host with the vulnerable NVIDIA driver
2. Locate and open the kernel-mode device interface exposed by the driver
3. Craft and submit input that triggers an oversized-to-smaller primitive cast
4. Observe truncation-induced corruption leading to driver crash or data exposure
5. Repeat or chain to amplify denial of service or harvest disclosed information

How to detect it

Review kernel logs for NVIDIA driver oops or truncation-related warnings, inspect process activity that opens GPU device nodes from non-root users, and search audit logs for unusual ioctl patterns against the driver.

What your security team should do

• Prioritize patching systems that expose NVIDIA GPUs to untrusted local users within the next maintenance window
• Audit local user access to GPU devices and restrict unnecessary privileges
• Monitor EPSS trend and any future KEV addition given the multi-year exposure window